CYP2C19 Genotype, Clopidogrel Metabolism, Platelet Function, and Cardiovascular Events:  A Systematic Review and Meta-analysis FREE

Clopidogrel is an antiplatelet drug used by approximately 40 million patients worldwide^1- 2 to treat or prevent atherothrombotic events and after percutaneous coronary revascularization. Quiz Ref IDAn overview of randomized trials including 7384 cardiovascular events in 79 613 patients with acute or stable coronary heart disease (CHD) or with multiple CHD risk factors demonstrated an association of clopidogrel therapy with reduced rates of cardiovascular events (odds ratio [OR], 0.88; 95% CI, 0.83-0.93) compared with placebo. Clopidogrel is also associated with a mechanism-based increase in major bleeding (OR, 1.28; 95% CI, 1.13-1.45).³ Despite the overall benefit, some individuals may be less responsive to clopidogrel than others⁴ because clopidogrel is a prodrug activated by several enzymes, including CYP2C19,⁵ and common genetic variation in CYP2C19 alters enzyme activity.^6- 7

In 2009, the US Food and Drug Administration (FDA) issued a boxed warning,⁸ recommending “consideration of CYP2C19 genotype” prior to prescribing clopidogrel. However, the American Heart Association and American College of Cardiologists argued there was insufficient evidence for this warning.⁹ Amid this uncertainty, FDA-cleared direct-to-consumer tests¹⁰ are available and demand is escalating.¹¹

We conducted a systematic review and critical appraisal to assess the strength and quality of evidence on the association of CYP2C19 genotype with responsiveness to clopidogrel.

Following guidance from the Human Genome Epidemiology Network (HuGEnet) on gene-disease association studies¹² and the reporting of systematic reviews from PRISMA,¹³ we searched PubMed and EMBASE from their inception to October 25, 2011. The search terms comprised the following¹⁴: drug name (including the generic name, clopidogrel, and trade names, eg, Plavix, clopilet, etc); metabolic enzymes (including CYP2C19); and gene (see the eMethods for full search strings) to identify studies describing associations of clopidogrel therapy with responsiveness (platelet response or clinical outcomes) in relation to CYP2C19 genotype (NCBI Entrez Gene 1557). Eligible articles had abstracts containing the keywords clopidogrel and CYP2C19 that reported original data (editorials and reviews were omitted). We interrogated bibliographies of included articles, identified previous meta-analyses and narrative reviews, and searched articles listed in the Pharmacogenomics Knowledge Base,¹⁵ the Medicines and Healthcare products Regulatory Agency (UK Department of Health), and the FDA. There was no language restriction.

Information extracted on 2 separate occasions included study design; follow-up duration; proportion of eligible individuals included in the pharmacogenetic analysis; inclusion criteria, eg, acute coronary syndrome (ACS), emergency or elective percutaneous intervention, or stable CHD; and the proportion of individuals receiving concomitant therapy with proton-pump inhibitors or aspirin. Uncertainties were resolved by contact with corresponding authors or by consensus. To evaluate risk of bias, we recorded reporting of outcome ascertainment, whether studies reported blinding to case status when ascertaining genotype, genotype indices (Hardy Weinberg equilibrium and call rate), and the source of funding.

The main analysis compared individuals with 1 or more copies of any CYP2C19 genetic variant associated with reduced enzyme function (ie, *2, *3, *4, *5, *6, *7, *8) with individuals categorized as having none of these alleles (*1/*1) or having 1 or more *17 gain-of-function alleles (the reference group). Separately, we conducted analyses of the association of 1 loss-of-function or 2 loss-of-function alleles with the same reference group. The primary cardiovascular outcome consisted of 1 or more of the following: all-cause mortality, fatal and nonfatal CHD, fatal and nonfatal stroke, stent thrombosis, target vessel revascularization, and hospitalization for ACS. Studies that reported stent thrombosis and no other outcome were excluded from the primary analysis. Additionally, we conducted separate analyses for the following end points: fatal and nonfatal myocardial infarction and stroke, stent thrombosis, and all-cause mortality. In studies that only reported hazard ratios, if provided we used the hazard proportion to estimate the number of events per genotype group. Otherwise, we substituted the relative risk (RR) for the summary hazards ratio. For studies that did not provide numerical values of effect estimates, we derived the point estimates and 95% confidence intervals from the published forest plots. We conducted similar analyses for the end points of any bleeding and major bleeding as defined by the contributing studies. For analysis of the association of CYP2C19 genotype and clopidogrel metabolites, we referred to the largest study by Mega et al.¹⁶ For studies of CYP2C19 genotype and platelet function, we noted the number of individuals and the mean (or median) and standard deviation (or interquartile range) for the relevant platelet function measure in different genotype categories.

For binary outcomes, we used RR as a measure of effect. We pooled studies using random- and fixed-effects models. Results reported in all figures and tables are from fixed-effects models unless otherwise stated. Sources of heterogeneity were investigated using meta-regression and quantified using the I² statistic.¹⁷ Tests for interaction using summary estimates were performed using the method described by Altman and Bland.¹⁸ For studies reporting platelet reactivity, we restricted the analysis to those with more than 500 participants and calculated standardized mean differences using fixed-effects (inverse variance) models. Small-study bias was assessed through the following: comparison of effect estimates in studies stratified by number of cardiovascular disease (CVD) events (1-99, 100-199, ≥200), the Harbord test, visual inspection of funnel plots, and trim-and-fill analysis. To evaluate the change in effect estimates with the addition of new evidence, we created cumulative meta-analysis plots. We estimated absolute risk differences between genotype categories based on the control group event rate from the CURE¹⁹ (ACS) and CHARISMA²⁰ (stable CHD) trials and the trim-and-fill summary RR (to minimize small-study bias), comparing carriers of any loss-of-function CYP2C19 allele vs the reference category, and assuming consistency of RRs across a range of baseline absolute risks of cardiovascular and bleeding events. We used a P value of <.05 to indicate statistical significance. All statistical tests were 2-sided. We used Stata version 11.2 for statistical analyses.

We identified 32 studies^16,21- 50 including 42 016 participants (29% female), with a weighted mean age of 63 years (Table, eTable 1, and eFigure 1). Twenty-one studies included patients with ACS at the time of recruitment and 8 included patients with stable CHD, mainly recruited at the time of coronary stent placement. (The remaining 3 studies did not report CHD status.) Of 32 studies, only 6 were nested within a randomized trial^{16,22,33,36,46}: in 5 of these, CYP2C19 genotyping was conducted in participants from both the clopidogrel and comparator groups. In 4 (referred to as “effect-modification” studies), the comparator was placebo (CURE, ACTIVE-A, CHARISMA, and CLARITY-TIMI 28^22,33,36). In the PLATO trial,⁴⁶ the comparator was ticagrelor. In the pharmacogenetic substudy of the TRITON-TIMI 38 trial,¹⁶ only the clopidogrel group was genotyped. All but 2 of the remaining 26 studies were prospective patient cohorts^{21,23- 27,29- 32,34- 35,37- 39,41- 45,47- 50} including participants who were all receiving clopidogrel (referred to as “treatment-only” studies). The remaining 2 were retrospective case-control studies.^28,40

Of the 28 CYP2C19 variant (*) alleles identified to date,⁵⁵ 13 were evaluated in the pharmacogenetic studies of clopidogrel (the loss-of-function *2 through *10 and *12 through *14 inclusive, as well as the gain-of-function *17) (Table). Individuals without a CYP2C19 variant associated with slow or fast metabolism were assumed to possess the *1 allele, indicating normal enzyme function. The *2 allele was typed by 31 of 32 studies (97%); both *3 and *17 alleles by 12 studies (38%), and the remaining 10 * alleles by 5 or fewer studies. Only 6 of the 28 single-nucleotide polymorphisms (SNPs) that uniquely identify the known CYP2C19 * alleles were listed in the Human HapMap (builds 21, 22, and 3[2]).⁵⁶ Only 2 studies reported detailed linkage disequilibrium between CYP2C19 SNPs,^28,38 and 1 study imputed an untyped SNP.³³

Twenty-six of 32 studies (81%) evaluated a composite outcome, but the individual components varied substantially between studies and included both hard clinical end points (eg, ST-segment elevation myocardial infarction) and softer end points (eg, hospitalization for ACS). A considerable proportion of disease outcomes were ascertained but not reported either individually or as part of the combined end point (eTable 2A and B).

Among individuals with the loss-of-function alleles *2 through *8 receiving a dose of 75 mg of clopidogrel (the usual maintenance dose), the mean area under the plasma concentration-time curve from the time of administration to the last measurable concentration (AUC_0-t) for the active clopidogrel metabolite was lower by 0.14 uM × hour compared with individuals carrying *1 or *17 alleles (Figure 1A). The difference in the active metabolite concentration between the 2 genotype categories was approximately half the geometric mean value seen in clopidogrel-treated individuals overall (AUC_0-t = 0.35 μM × hour). In 4 studies using the treatment-only approach (4341 individuals)^{40,45,57- 58} in which the intermediate phenotype of platelet reactivity was assessed following 600 mg of clopidogrel, there was a per-allele association of the *2 variant on platelet aggregation when compared with subjects with *1*1 genotype (Figure 1B).

A pooled analysis of 22 studies using the treatment-only approach,^{16,21,24- 27,29- 32,34- 35,37- 38,41- 45,47- 48,50} supplemented by a treatment-only analysis using data from the clopidogrel-treatment group of 4 randomized trials^22,36,46 (with a total 2465 clinical events among 26 251 individuals), indicated that individuals with any copy of CYP2C19 alleles *2 through *8 when compared with individuals with *1 or *17 alleles had a higher risk of CVD events (RR, 1.18; 95% CI, 1.09-1.28; I² = 60%; 95% CI, 38%-75%, using fixed-effects models and RR, 1.34; 95% CI, 1.15-1.56, using random-effects models) (Figure 2). Meta-cumulative analysis, with the genotype group reversed to make the comparison akin to a trial of a higher vs lower dose of active clopidogrel metabolite, showed that at the time of FDA approval, individuals with *1 or *17 had an RR of CVD events of 0.82 (95% CI, 0.72-0.93) for fixed-effects and an RR of 0.72 (95% CI, 0.57-0.92) for random-effects modeling compared with individuals with *2 through *8, with some attenuation of the summary estimate with addition of new studies (most recent estimate: RR, 0.86; 95% CI, 0.79-0.94, for fixed-effects and RR, 0.76; 95% CI, 0.65-0.89, for random-effects modeling) (Figure 3).

When studies were stratified by number of outcome events, there was a clear trend toward the null in larger studies (Figure 2), consistent with small-study bias (χ²₂ = 30; P = 3.2 × 10⁻⁷). The Harbord test for small-study bias was positive (P = .001), and the funnel plot was asymmetric (eFigure 2). When we quantified the potential effect of small-study bias using trim-and-fill analysis, addition of the 8 hypothetical missing studies reduced the summary RR to 1.10 (95% CI, 1.02-1.19) for fixed-effects and an RR of 1.13 (95% CI, 0.96-1.33) for random-effects models. In the 4 largest studies reporting 200 or more CVD events (51% of all reported events), the summary RR was 0.97 (95% CI, 0.86-1.09) for fixed-effects and 0.95 (95% CI, 0.81-1.11) for random-effects modeling. Assuming a control event rate of 73 and 114 CV events per 1000 individuals in the setting of stable CHD and ACS, respectively, the estimated excess number of CV events for subjects carrying *2 to *8 alleles ranged from 8 to 12 events per 1000 individuals (eTable 3).

Eleven of 32 studies involving 10 291 individuals provided data that allowed us to obtain the effect of 1 loss-of-function allele (>238 CVD events) or 2 loss-of-function alleles (>37 CVD events) compared with the *1 or *17 reference group. The RR of CVD events among carriers of 1 loss-of-function allele was 1.77 (95% CI, 1.27-2.47; I² = 45%; 95% CI, 0%-78%) for fixed-effects and 2.01 (95% CI, 1.21-3.34) for random-effects modeling in studies with fewer than 100 cases. In studies with 100 or more events, the values for the same exposure were an RR of 0.94 (95% CI, 0.80-1.10; I² = 31%; 95% CI, 0%-75%) for fixed-effects and an RR of 0.95 (95% CI, 0.78-1.15) for random-effects modeling. Using the estimate from studies reporting 100 or more cases, this corresponded to 4 to 7 fewer CVD events per 1000 individuals in the setting of stable CHD and ACS, respectively. By comparison, the RR of CVD events among carriers of 2 loss-of-function alleles was 3.75 (95% CI, 2.40-5.86; I² = 8%; 95% CI, 0%-73%) for fixed effects and 3.76 (95% CI, 2.34-6.06) for random-effects modeling in studies with fewer than 100 cases and 1.52 (95% CI, 1.04-2.21; I² = 51%; 95% CI, 0%-86%) for fixed effects and 1.45 (95% CI, 0.82-2.56) for random-effects modeling in studies with 100 or more cases (eFigure 3). Using the estimate from studies reporting 100 or more cases, this corresponded to an absolute increase of 38 and 59 CVD events per 1000 individuals in the setting of stable CHD and ACS, respectively.

Stent thrombosis was the clinical outcome with the strongest association with CYP2C19 loss-of-function alleles (579 cases in 16 008 individuals: RR, 1.75; 95% CI, 1.50-2.03; I² = 44%; 95% CI, 0%-70% for fixed-effects and RR, 1.88; 95% CI, 1.46-2.41 for random-effects modeling), but a trend toward the null was observed in larger studies (Figure 4). Assuming an event rate of 18 per 1000 individuals in the control group,⁵⁹ this corresponded to an absolute increase of 14 stent thromboses per 1000 individuals.

Individuals carrying any copy of *2 through *8 alleles had a lower risk of any bleeding when compared with individuals carrying the *1 or *17 allele (1025 events in 4182 individuals; RR, 0.84; 95% CI, 0.75-0.94; I² = 0%; 95% CI, 0%-89% for both fixed- and random-effects modeling) (Figure 4), corresponding to an absolute reduction in the risk of bleeding that ranged from 5 to 8 events per 1000 individuals depending on clinical setting (and assuming a control event rate of 31 and 50 bleeding events per 1000 individuals for stable CHD and ACS, respectively) (eTable 3). Other clinical outcomes (including all-cause mortality, myocardial infarction, and stroke) with respective fixed- and random-effects estimates are presented in Figure 4 and eTable 4A.

There was no evidence that CHD status at study recruitment (ACS vs stable CHD), blinding to genotype, source of funding, and concomitant use of proton-pump inhibitors or aspirin modified the association of CYP2C19 genotype with CVD events (P  .09 for all meta-regressions) (eFigure 4).

In the 4 placebo-controlled randomized trials^22,33,36 (11 477 subjects and >1097 major CVD events), the RR of major CVD events in patients treated with clopidogrel compared with placebo was 0.78 (95% CI, 0.69-0.89) in genotype category *1 or *17 and 0.87 (95% CI, 0.70-1.09) in genotype category *2 or *3 (z = 0.89; P = .37 for interaction) (Figure 5A). We found weak evidence for a treatment × genotype interaction for major bleeding; the RR for major bleeding comparing clopidogrel with placebo was 1.99 (95% CI, 1.31-3.02) in genotype category *2 or *3 and 1.28 (95% CI, 1.02-1.61) in genotype category *1 or *17 (z = 1.83; P = .07 for interaction) (Figure 5B), but the pattern was inconsistent with that predicted from the relationship between genotype and active clopidogrel metabolite concentration or platelet function (Figure 1). The evidence remained similar when using random-effects modeling (eTable 4B).

Quiz Ref IDDespite associations between CYP2C19 genotype, clopidogrel metabolism, and platelet aggregation, this systematic review and meta-analysis does not demonstrate a clinically important association of genotype with cardiovascular outcomes with the possible exception of stent thrombosis. Treatment-only studies were affected by small-study bias while the effect-modification studies provided no evidence for a genotype × treatment interaction. Our appraisal has identified issues pertaining to the design and analysis of pharmacogenetic studies that are important in the understanding of the significance of the CYP2C19 genotype.

Quiz Ref IDOnly 5 studies of clopidogrel treatment response were nested within a randomized trial in which both treatment and comparator group were genotyped, a design that allows evaluation of the clopidogrel treatment effect overall (through comparison of event rates in treatment and control groups) and an assessment of differential treatment response by CYP2C19 genotype (assessed as a statistical test for genotype × treatment interaction). Randomized trials are routinely registered in advance of reporting, and trial protocols and study outcomes are usually prespecified, reducing the risk of publication and selective reporting bias. Among the 4 effect-modification placebo-controlled trials included in this meta-analysis, there was no evidence of a clopidogrel treatment × CYP2C19 genotype interaction.

By contrast, most studies examining the relationship between CYP2C19 genotype and cardiovascular outcomes were performed among patients treated with clopidogrel, usually in an observational setting. In these treatment-only studies, subjects classified as poor metabolizers (ie, carrying any *2 to *8 CYP2C19 allele) had a higher risk of CVD when compared with normal or fast metabolizers (*1 or *17). However, there was strong evidence of small-study bias. We also identified evidence for selective outcome reporting bias. Not all ascertained outcomes were reported either individually or as part of the CVD composite (eTable 2A and B). Moreover, if a particular * allele was not typed, the participant was presumed to lack the variant allele in that position. Thus, some individuals assigned to the *1 category (lacking all alternative alleles) may have been misclassified, simply because genotyping was not attempted for some * alleles.⁶⁰ Taken together, these observations cast doubt on the association of CYP2C19 genotype with clinical cardiovascular end points.

Aside from biases that may have influenced treatment-only studies, some other design limitations are noteworthy. Inferring effect modification through a treatment-only analysis would be considered invalid for a nongenetic variable (age or sex), but there are 3 reasons for its widespread use (~95%)⁶¹ in pharmacogenetic studies. First, ethical, logistic, or financial considerations may preclude DNA collection from trial participants. Second, based on Mendel's second law of independent assortment, random allocation of genetic variants from parent to offspring should mean that groups categorized by genotype for a drug metabolizing enzyme do not differ systematically (except in their response to drug treatment).⁶² Third, it is implicitly assumed that genetic variation in a drug metabolizing enzyme should have no disease association in the absence of exposure to medication.⁶³ Although the CYP2C19 genotype has not been associated with CHD events in individuals not treated with clopidogrel, CYP2C19 genotype was associated with inflammation markers (linked to CVD risk) in a candidate gene study,⁶⁴ and recent genome-wide association studies have implicated other CYP gene variants in the susceptibility to elevated blood pressure (CYP1A1,⁶⁵CYP1A2,⁶⁵ and CYP17A1^65- 66) and CHD (CYP17A1⁶⁷).

If Mendel's second law and Motulsky's assumptions⁶³ apply, a treatment-only study of a drug-metabolizing gene variant can be considered analagous to a randomized trial of a high vs low dose of a drug with the association with clinical outcomes being related to the difference in concentration of the active metabolite between the genotype categories for a given administered dose. The summary RR in CVD events seen in placebo-controlled trials of 75-mg clopidogrel, irrespective of genotype, was 0.88 (95% CI, 0.83-0.93). In the study by Mega et al,¹⁶ the ratio of the difference in mean AUC values for the active clopidogrel metabolite for the 2 major CYP2C19 genotype categories (*2 to *8 vs *1 or *17) to the mean AUC regardless of genotype was 0.14/0.35 (Figure 1A). Assuming a log-concentration response relationship between the active clopidogrel metabolite and CVD events, the RR in CVD events comparing individuals with 1 or more CYP2C19 gain-of-function alleles with the reference category should be 0.95 (ie, 0.88^[0.14/0.35]), with the genotype comparison switched as compared with most of the genotype comparisons in the “Results” section, to make it akin to a trial of a higher vs lower dose of active clopidogrel metabolite. However, the observed summary estimate for the association of the gain-of-function genotype with CVD events in the treatment-only studies available at the time of the FDA boxed warning suggested greater benefit from CYP2C19 than predicted (RR, 0.82; 95% CI, 0.72-0.93) (Figure 3). With the addition of 13 studies, the benefit associated with the gain-of-function genotype is slightly less (RR, 0.86; 95% CI, 0.79-0.94) yet still suggests greater benefit than the predicted effect (RR, 0.95) and is similar in magnitude to the RR observed in the meta-analysis of trials of 75-mg clopidogrel vs placebo, regardless of genotype.

Two other observations raise questions regarding the utility of CYP2C19 genotyping in evaluation of clopidogrel response. Where the major harms of a drug treatment are mechanism-based, as is the case for clopidogrel where the major adverse effect is bleeding, any attempt to individualize dose to optimize treatment response (ie, to reduce CVD risk) is likely to be offset by an opposing effect on the rate of harm (ie, risk of bleeding), and this was observed in the current analysis (eTable 3). The net potential clinical benefit of genotyping to adjust dose may therefore not be as great as initially assumed. Quiz Ref IDWhen genetic variants under investigation lie in drug metabolizing enzymes, the interaction will likely be quantitative (ie, differences in the magnitude of effect) rather than qualitative (differences in the direction of effect)⁶⁸; thus, even if a “real” difference between subgroups is detected, the net clinical utility could be limited (because the drug could still be beneficial in individuals with the “poor” responder genotype), and a careful cost-effectiveness analysis is necessary to justify a change in clinical practice.

Quiz Ref IDOur interpretation is consistent with the findings of a recent meta-analysis,⁶⁹ but the current review differs from that one in several important ways. First, the current review includes 17 more studies than the recent meta-analysis. Second, the current review includes analysis of the association between the CYPC19 genotype and clopidogrel metabolites and platelet reactivity. Third, the current review includes analysis of individual outcomes as well as composite cardiovascular outcomes. Fourth, the current review compares and contrasts findings from treatment-only and effect-modification study designs. Fifth, we placed the observed risk of cardiovascular outcomes in the context of the expected risk based on the observed effect of CYP2C19 genotype on active clopidogrel metabolites.

Our study has several limitations. We used aggregate (not participant-level) data and our power to detect differences in subgroups (eg, based on use of proton-pump inhibitors or aspirin) was consequently limited. The components of the composite CVD end points differed across studies, which may have influenced the meta-analysis summary RR.^70- 71 However, on sensitivity analysis, the summary RR was not overly influenced by the exclusion of individual studies (eFigure 5). Moreover, we also examined the main components of the composite end point. The inclusion of studies with both stable CHD and ACS might dilute an association if the magnitude of effect of clopidogrel was greater in ACS than in patients with stable CHD; however, when we stratified studies according to recruitment at the time of ACS, we found no heterogeneity in the association between CYP2C19 genotype and risk of CVD events (eFigure 4). We report estimates from both fixed- and random-effects meta-analysis modeling because of considerable between-study heterogeneity in the effect of CYP2C19 on risk of CVD events in treatment-only analysis (I² = 60%). However, when adjusting for number of events per study, the I² attenuated to 20%, a value considered to reflect low heterogeneity.⁷² Therefore, by assigning more weight to smaller studies,^73- 74 the random-effects model may exacerbate the small-study bias identified in our analysis.

In conclusion, this study identified no clinically significant interaction of CYP2C19 genotype with the association of clopidogrel therapy and cardiovascular events.